There is no magic in there. What NMA does is it finds (orthogonal) correlated displacements that are associated with lowest change in potential energy.

Obviously, in absence of external fields translation and rotation should not require any energy input, so the corresponding 6 eigenvalues are 0 (6 because there are 6 translational/rotational modes available that are orthogonal to each other, i.e., their pairwise scalar products are all equal to 0).

Further modes are just those that require the least energy to "activate". Since global conformational changes are driven by thermal fluctuations or minor perturbations due to e.g. ligand binding, we assume they *have* to be low energy (and thus low frequency) modes. For that reason, there is a good chance that low-energy modes (or, in a more pictorial manner, broad potential energy basins) coincide with relevant conformational transitions.

This approach is of course limited by considering only a single static structure. If you have an actual equilibrium trajectory in a finite temperature, you could obtain much more reliable data from PCA, which instead looks for largest collective motions of atoms by considering covariance/correlation between atomic positions.

It is an old (years 1980s) idea based on the quasi-harmonic approximation introduced by Karplus and coworkers (Karplus and Kushick,

Macromolecules vol. 14, page 325 (1981)) and refined over the years by other groups. A good review with all the original references I propose is Protein Engineering

vol.14 no.1 pages 1–6 (2001). In short, the basic idea is that within the normal mode theory the lowest frequency  compares rather well with the conformational change observed by crystallographers in some synthases and hexokinases. Yet, there is no clear guarantee of generality.

There is no magic in there. What NMA does is it finds (orthogonal) correlated displacements that are associated with lowest change in potential energy.

Obviously, in absence of external fields translation and rotation should not require any energy input, so the corresponding 6 eigenvalues are 0 (6 because there are 6 translational/rotational modes available that are orthogonal to each other, i.e., their pairwise scalar products are all equal to 0).

Further modes are just those that require the least energy to "activate". Since global conformational changes are driven by thermal fluctuations or minor perturbations due to e.g. ligand binding, we assume they *have* to be low energy (and thus low frequency) modes. For that reason, there is a good chance that low-energy modes (or, in a more pictorial manner, broad potential energy basins) coincide with relevant conformational transitions.

This approach is of course limited by considering only a single static structure. If you have an actual equilibrium trajectory in a finite temperature, you could obtain much more reliable data from PCA, which instead looks for largest collective motions of atoms by considering covariance/correlation between atomic positions.